Balaji Arumugam
Portfolio of Balaji Arumugam - Senior Developer, Indie Hacker, and AI enthusiast
Using Claude in Github: Hype and Reality
Agents are coming faster than ever and yet are still kind of hard to use?
How we decreased GitLab repo backup times from 48 hours to 41 minutes
Learn how GitLab tracked a performance bottleneck to a 15-year-old Git function and fixed it, leading to enhanced efficiency that supports more robust backup strategies and can reduce risk.
Sharing everything I could understand about gradient noise
You've most likely heard about gradient noise through the name Perlin noise, which refers to one particular implementation with various CPU optimizations. Because it's an incredible tool for creative work, it's used virtually everywhere: visual effects, video …
I can’t believe they said that! How to speak difficult truths
Why comedians can say things other people can't
Have LLMs Finally Mastered Geolocation? - bellingcat
We tasked LLMs from OpenAI, Google, Anthropic, Mistral and xAI to geolocate our unpublished holiday snaps. Here's how they did.
Experts in the Internet Age: the Power of Email
These days I spend a fair bit of my time emailing libraries in order to get access to research materials for a project I'm working on, and I'd like to share what has become a very typical experience for me and one that I think more people should know about.
Vibe Coding: Where it works and where it doesn't
Vibe coding is a term that was coined by Andrej Karpathy in a tweet and it has been catching fire since then every day.
Jason Eckert's Website and Blog
My personal website and blog
MCP Resources Are For Caching
This is a quick tour of what MCP resources actually are. And more to the point, what MCP is supposed to do (and not supposed to do).
Why is it hard to evaluate GenAI applications? | Random Walk
TL;DR If you don’t have time to read the whole article, the following four takeaways are a concise version. You can navigate to the corresponding section in the article for details. Lack of framework: A GenAI application is not a GenAI foundation model; diffe…
Yet Another Dating App (some ideas)
Ramblings on a new kind of dating app
Tip: Put your Rails app on a SQL query diet
Introduction Much of the time taken processing HTTP requests in web apps is SQL queries. To minimize that, we want to avoid unnecessary and duplicate queries, and generally perform as few queries as possible.
Working on the plumbing in a small web community :: Sacha Chua
The IndieWeb Carnival prompt for May is small web communities. I've been exploring some thoughts on how a little effort goes a long way to connecting a community. Sometimes I think of it as working on the plumbing so that ideas can flow more smoothly. It feels…
Writing is fun again (with LLMs)
After I started FunnelStory, I just stopped writing. As a founder, you always have a list of ten other things that need to get done, and writing never felt like the most important one. It took up too much time. Ever since I started using LLMs for writing, thou…
A bored engineer's personal website
A place to ramble about my electrical engineering classes, personal hardware projects and research notes.
Highly efficient matrix transpose in Mojo 🔥
In this blogpost I will step by step show you how to implement a highly efficient transpose kernel for the architecture using Mojo. The best kernel archive...
Meta Is Turning Private AI Chats Into Public Content | Demand Privacy Protections
Meta’s new AI app is making private conversations public without clear consent. Join the Mozilla community in demanding Meta shut down its "Discover Feed" and ensure that all AI interactions are private by default. Protect user privacy now.
I Do Not Remember My Life and It's Fine
What reminiscing is like without mental imagery
Eleven v3 (alpha) — The most expressive Text to Speech model
Generate lifelike speech in 70+ languages with emotion, direction, and multi-speaker control using inline audio tags.
Digital Minister wants open standards and open source as guiding principle
At re:publica, Digital Minister Wildberger advocates for more sovereignty and calls for alternatives to digital payments and social media.
Self-hosting your own media considered harmful
I just received my second community guidelines violation for my video demonstrating the use of LibreELEC on a Raspberry Pi 5, for 4K video playback.
Decrease in Entry-Level Tech Jobs
The percentage of hires for entry-level tech jobs has drastically decreased. This is what you need to do to increase your chances of getting hired.
macOS Wishlist: Gestures
Wherein I make a video to show off the concept of gestures for macOS.
The confluence
Britain's problems are all compounding one another. We need a total change of direction.
Analyzing Metastable Failures in Distributed Systems
So it goes: your system is purring like a tiger, devouring requests, until, without warning, it slumps into existential dread. Not a crash. ...
Learning Elvish (but not the Middle-earth one)
Asocial programmer's blog
Generating Pixels One by One
Your First Autoregressive Image Generation Model
How to improve data quality
Using automated systems to detect errors in data
The Compound Error Crisis: Why LLM Agents Are Failing Like Broken Robots (And Why Computer Science Warned Us)
A 6-axis robot arm reaches for a coffee cup. Joint 1 is off by 0.5 degrees. Joint 2 compensates but overshoots by 0.8 degrees. By the time the arm reaches the cup, it's 3 inches to the left and...
Benchmarking is hard, sometimes ...
Benchmarking complex systems can be difficult. Here's a problem I'm currently investigating, possibly related to process scheduling.
Process improvement is more important than metrics
process improvement is more important than metrics
Three-Tier Storage Architecture for Fast LLM Inference in the Cloud
Large Language Model (LLM) inference workloads deal with extremely large model files (often many gigabytes) that must be loaded quickly and repeatedly across distributed GPU instances. Traditional single-tier storage (either solely local disk or only remote cl…
Essay-driven landing pages
There is a pretty unconventional way to build landing pages – totally different from how like 99% of people do it nowadays.
Extracting SSH private keys from Claude training data
June 5, 2025
The ethics of README ads
I’ve been considering accepting sponsorship again for my projects.
Why more `/sys/class/hwmon/*/temp*_label` than CPU cores?
So, I am currently working on a Linux tool+daemon to replace auto-cpufreq with a more efficient, tunable and observable alternative.